예상 투표 분포는 정당 정체성에 따라 달라진다

A Expected vote within each party groupB Deviation from independenceClintonDoleDemocratn = 48895.7%467 / E 284.8res +24.121 / E 203.2res -24.1Independentn = 3770.3%26 / E 21.6res +1.529.7%11 / E 15.4res -1.5Republicann = 41958 / E 244.6res -24.886.2%361 / E 174.4res +24.8ClintonDoleχ²(2) = 623.8 · p < .001V 0.813 [0.776, 0.849]

3개 정당군 · Counts are observed. E = expected counts. Residual color is descriptive; no cellwise significance claim.

ANES96 / statsmodels public-domain subset ·944respondents · unweighted exploratory reanalysis

01 / 18χ² 독립성 · 1/2

01 / 18χ² 독립성 · 1/2

원래7수준에서도 정당 정체성과 예상 투표의 연관은 유지된다

A Expected vote within each party groupB Deviation from independenceClintonDole0 Strong Dem.n = 20098.5%197 / E 116.7res +133 / E 83.3res -131 Weak Dem.n = 18093.9%169 / E 105.1res +10.711 / E 74.9res -10.72 Lean Dem.n = 10893.5%101 / E 63res +7.97 / E 45res -7.93 Independentn = 3770.3%26 / E 21.6res +1.529.7%11 / E 15.4res -1.54 Lean Rep.n = 9425.5%24 / E 54.9res -6.874.5%70 / E 39.1res +6.85 Weak Rep.n = 15026 / E 87.6res -11.182.7%124 / E 62.4res +11.16 Strong Rep.n = 1758 / E 102.1res -1695.4%167 / E 72.9res +16ClintonDoleχ²(6) = 637.2 · p < .001V 0.822 [0.79, 0.856]

원래7수준 · Counts are observed. E = expected counts. Residual color is descriptive; no cellwise significance claim.

ANES96 / statsmodels public-domain subset ·944respondents · unweighted exploratory reanalysis

02 / 18χ² 독립성 · 2/2

02 / 18χ² 독립성 · 2/2

예상 투표집단의 평균연령 차이는 구간으로 판단한다

A Group means · pointwise 95% CI43.9345.6247.3249.0250.72Clinton 예상46.3[44.93, 47.67]Dole 예상48.09[46.46, 49.72]yearsB Difference+1.79years95% CI[-0.34, 3.914]t(843.5) = 1.65p = 0.1Hedges g = 0.109effect CI [-0.029, 0.237]Group n: 551 / 393 · independent groups

독립 Welch 비교 · Mean age differs by expected vote; Welch df reflects unequal variances. Nonrandom groups.

ANES96 / statsmodels public-domain subset ·944respondents · unweighted exploratory reanalysis

03 / 18t-test · 1/2

03 / 18t-test · 1/2

같은 응답자의 두 후보 인식에는 대응 차이가 있다

A Group means · pointwise 95% CI12.545.57Clinton 인식2.94[2.85, 3.03]Dole 인식5.39[5.31, 5.48]ideology-scale pointsB Difference+2.45ideology-scale points95% CI[2.315, 2.594]t(943) = 34.54p < .001Cohen dz = 1.124effect CI [1.014, 1.249]Group n: 944 / 944 · same 944 respondentsScale: 1 very liberal → 7 very conservative

대응 인식 비교 · DoleLR−ClinLR on the same 1–7 numeric-coded scale. Not a causal candidate effect.

ANES96 / statsmodels public-domain subset ·944respondents · unweighted exploratory reanalysis

04 / 18t-test · 2/2

04 / 18t-test · 2/2

연령과 뉴스 이용은 함께 움직이지만 인과를 뜻하지 않는다

A Observed age × TV-news frequencyB Pearson 수치 correlationr [95% CI]024620406080Age (years)TV news (times/week)AgeAge10.41[0.35, 0.46]-0.16[-0.22, -0.1]-0.02[-0.09, 0.04]NewsNews0.41[0.35, 0.46]1-0.06[-0.12, 5.72e-3]-1.58e-3[-0.07, 0.06]Educ.Educ.-0.16[-0.22, -0.1]-0.06[-0.12, 5.72e-3]1-0.02[-0.08, 0.05]UrbanUrban-0.02[-0.09, 0.04]-1.58e-3[-0.07, 0.06]-0.02[-0.08, 0.05]1All pairs n = 944 · point area = observed countNo jitter or fabricated observations. Correlation does not identify a causal relation.

Pearson 수치 · Pearson intervals use Fisher-z. Numeric-coded education is not an equal-interval measurement.

ANES96 / statsmodels public-domain subset ·944respondents · unweighted exploratory reanalysis

05 / 18Correlation · 1/2

05 / 18Correlation · 1/2

순위로 바꾸어도 관계의 방향과 불확실성을 함께 읽는다

A Observed age × TV-news frequencyB Spearman 순위 correlationρ [95% CI]024620406080Age (years)TV news (times/week)AgeAge10.4[0.35, 0.45]-0.13[-0.2, -0.07]-0.05[-0.12, 8.59e-3]NewsNews0.4[0.35, 0.45]1-0.06[-0.13, 5.82e-3]-5.80e-3[-0.07, 0.06]Educ.Educ.-0.13[-0.2, -0.07]-0.06[-0.13, 5.82e-3]1-0.02[-0.08, 0.04]UrbanUrban-0.05[-0.12, 8.59e-3]-5.80e-3[-0.07, 0.06]-0.02[-0.08, 0.04]1All pairs n = 944 · point area = observed countNo jitter or fabricated observations. Correlation does not identify a causal relation.

Spearman 순위 · Ordinal sensitivity:rank-pair percentile bootstrap, 2000 draws. Matrix intervals are pointwise,not simultaneous.

ANES96 / statsmodels public-domain subset ·944respondents · unweighted exploratory reanalysis

06 / 18Correlation · 2/2

06 / 18Correlation · 2/2

다른 변수를 함께 고려해도 연령과 뉴스 이용은 연관된다

A Unstandardized B ·95% CI-1.31-0.75-0.190.370.92Independent vs Dem.-0.33[-1.15, 0.49]Republican vs Dem.-0.27[-0.6, 0.05]Age (+10 yr)+0.67[0.58, 0.77]Education code+0.02[-0.08, 0.12]Urban log(pop)+2.49e-4[-0.05, 0.05]Times/week per term unitB Adjusted mean response23456720406080Age (years)TV news (times/week)R² = 0.17 · N =944

고전적 표준오차 · Age: 10 years; education: 1–7 code. Urban:log(population in1000s+.1). Other covariates averaged over this sample; band is a mean CI.

ANES96 / statsmodels public-domain subset ·944respondents · unweighted exploratory reanalysis

07 / 18Regression · 1/2

07 / 18Regression · 1/2

오차 가정을 바꾸면 계수보다 신뢰구간이 달라진다

A Unstandardized B ·95% CI-1.57-0.94-0.320.310.93Independent vs Dem.-0.33[-1.4, 0.73]Republican vs Dem.-0.27[-0.59, 0.04]Age (+10 yr)+0.67[0.58, 0.76]Education code+0.02[-0.08, 0.12]Urban log(pop)+2.49e-4[-0.05, 0.05]Times/week per term unitB Adjusted mean response23456720406080Age (years)TV news (times/week)R² = 0.17 · N =944

HC3 강건 오차 · Age: 10 years; education: 1–7 code. Urban:log(population in1000s+.1). Other covariates averaged over this sample; band is a mean CI.

ANES96 / statsmodels public-domain subset ·944respondents · unweighted exploratory reanalysis

08 / 18Regression · 2/2

08 / 18Regression · 2/2

교육집단 평균 차이의 증거는 뚜렷하지 않다

A Actual0–7 times distributionsB Familywise mean contrastsHigh school or lessn 313 · mean 3.9301234567Some collegen 187 · mean 3.6501234567Degree or moren 444 · mean 3.6201234567-0.97-0.57-0.180.220.622 − 1-0.27[-0.86, 0.31]3 − 1-0.31[-0.78, 0.16]3 − 2-0.03[-0.57, 0.51]Times/week · corrected95% CIF(2, 481.8) = 1.25p = 0.287 · descriptive η² 0.0028

Welch + Games–Howell · Groups: 1 = high school or less; 2 = some college; 3 = degree+. Bars show within-group frequencies; contrasts use the selected correction.

ANES96 / statsmodels public-domain subset ·944respondents · unweighted exploratory reanalysis

09 / 18One-way ANOVA · 1/2

09 / 18One-way ANOVA · 1/2

분산 가정과 쌍별 보정법을 함께 제시해야 한다

A Actual0–7 times distributionsB Familywise mean contrastsHigh school or lessn 313 · mean 3.9301234567Some collegen 187 · mean 3.6501234567Degree or moren 444 · mean 3.6201234567-0.96-0.57-0.170.230.622 − 1-0.27[-0.85, 0.31]3 − 1-0.31[-0.77, 0.16]3 − 2-0.03[-0.58, 0.51]Times/week · corrected95% CIF(2, 941) = 1.301p = 0.273 · descriptive η² 0.0028

고전적 F + Tukey · Groups: 1 = high school or less; 2 = some college; 3 = degree+. Bars show within-group frequencies; contrasts use the selected correction.

ANES96 / statsmodels public-domain subset ·944respondents · unweighted exploratory reanalysis

10 / 18One-way ANOVA · 2/2

10 / 18One-way ANOVA · 2/2

불균형한9개 셀의 평균은 표본 수와 함께 읽는다

A Full-interaction cell means ·95% CIB Type III F tests02468Age bandTV news (times/week)18–3435–5455+Democratn: 131 / 210 / 147Independentn: 6 / 22 / 9Republicann: 100 / 181 / 138Party groupF(2, 935) = 1.329p = 0.265Age bandF(2, 935) = 17.673p < .001Party × ageF(4, 935) = 1.075p = 0.367

Type III F · Unbalanced 3×3 design, sum contrasts. Cell n and pointwise intervals remain visible. Joint interaction test is not inferred from line crossings.

ANES96 / statsmodels public-domain subset ·944respondents · unweighted exploratory reanalysis

11 / 18Two-way ANOVA · 1/2

11 / 18Two-way ANOVA · 1/2

선의 교차보다 상호작용의 공동 검정을 먼저 본다

A Full-interaction cell means ·95% CIB HC3 joint F tests02468Age bandTV news (times/week)18–3435–5455+Democratn: 131 / 210 / 147Independentn: 6 / 22 / 9Republicann: 100 / 181 / 138Party groupF(2, 935) = 1.312p = 0.27Age bandF(2, 935) = 11.457p < .001Party × ageF(4, 935) = 0.757p = 0.553

HC3 joint F · Unbalanced 3×3 design, sum contrasts. Cell n and pointwise intervals remain visible. Joint interaction test is not inferred from line crossings.

ANES96 / statsmodels public-domain subset ·944respondents · unweighted exploratory reanalysis

12 / 18Two-way ANOVA · 2/2

12 / 18Two-way ANOVA · 2/2

연령의 조건부 연관은 전체 상호작용 검정과 구분한다

A Conditional mean responseB Conditional age slope ·95% CI23456720406080Age (years)TV news (times/week)00.20.40.6048Urban log(population in1000s+.1)Times/week per10yearsUrban P10+0.759 [0.57, 0.948]Urban P50+0.667 [0.569, 0.764]Urban P90+0.613 [0.466, 0.761]Interaction B -0.017 · 95% CI [-0.049, 0.015] · p = 0.298

고전적 조건부 효과 · Observed moderator range; P10/P50/P90 lines; dashed bounds show each mean CI. Intervals retain full covariance and are pointwise,not simultaneous.

ANES96 / statsmodels public-domain subset ·944respondents · unweighted exploratory reanalysis

13 / 18Interaction / Moderation · 1/2

13 / 18Interaction / Moderation · 1/2

강건한 오차에서도 상호작용 구간은0을 포함한다

A Conditional mean responseB Conditional age slope ·95% CI23456720406080Age (years)TV news (times/week)00.20.40.6048Urban log(population in1000s+.1)Times/week per10yearsUrban P10+0.759 [0.567, 0.951]Urban P50+0.667 [0.576, 0.757]Urban P90+0.613 [0.46, 0.767]Interaction B -0.017 · 95% CI [-0.052, 0.018] · p = 0.336

HC3 조건부 효과 · Observed moderator range; P10/P50/P90 lines; dashed bounds show each mean CI. Intervals retain full covariance and are pointwise,not simultaneous.

ANES96 / statsmodels public-domain subset ·944respondents · unweighted exploratory reanalysis

14 / 18Interaction / Moderation · 2/2

14 / 18Interaction / Moderation · 2/2

모형의 간접 연결은 인과적 매개를 입증하지 않는다

A Structural path overviewB Direct,indirect and total associationIndustrialization1960Democracy1960Democracy1965a = 1.483b = 0.837c′ = 0.572Indirect a×b = 1.242Same measurement + residual modelN = 75 countries · MLW fit-0.20.541.282.022.77간접 a×b+1.24[0.57, 1.97]직접 c′+0.57[0.14, 1.01]총 c′+a×b+1.81[1.09, 2.56]Unstandardized latent-scale B

Monte Carlo 구간 · Parametric Monte Carlo 200,000 draws from full parameter covariance; not a nonparametric bootstrap. Observational indirect association.

Political Democracy ·75countries · semopy2.3.11 / lavaan official example · exploratory model

15 / 18Mediation · 1/2

15 / 18Mediation · 1/2

간접 연결의 구간은 계산법과 공분산에 의존한다

A Structural path overviewB Direct,indirect and total associationIndustrialization1960Democracy1960Democracy1965a = 1.483b = 0.837c′ = 0.572Indirect a×b = 1.242Same measurement + residual modelN = 75 countries · MLW fit-0.20.531.272.012.75간접 a×b+1.24[0.55, 1.94]직접 c′+0.57[0.14, 1.01]총 c′+a×b+1.81[1.08, 2.55]Unstandardized latent-scale B

Delta-method 구간 · First-order delta normal intervals retain cross-parameter covariance. Path direction is model structure,not causal proof.

Political Democracy ·75countries · semopy2.3.11 / lavaan official example · exploratory model

16 / 18Mediation · 2/2

16 / 18Mediation · 2/2

측정·구조·잔차를 구분하면 복잡한 모형도 읽을 수 있다

Industrialization1960x1 · GNP/capita1εx2 · Energy2.18εx3 · Industry1.82εDemocracy1960y1 · Press1εy2 · Opposition1.26εy3 · Elections1.06εy4 · Legislature1.26εDemocracy1965y5 · Press1εy6 · Opposition1.19εy7 · Elections1.28εy8 · Legislature1.27ε1.483[0.7, 2.27]0.837[0.64, 1.03]Direct 0.572 [0.14, 1.01]0.621.312.150.790.351.36χ²(35) 38.13 · p = 0.329CFI 0.995 · RMSEA 0.035B unstandardized · first loadings fixed at 1Dashed: residual covariance · ε: indicator error

잔차 공분산6개 포함 · 75 countries, 11 indicators, 3 latent variables. Structural path CIs are normal expected-Fisher intervals. Good fit is not causal identification.

Political Democracy ·75countries · semopy2.3.11 / lavaan official example · exploratory model

17 / 18SEM · 1/2

17 / 18SEM · 1/2

잔차 공분산을 생략하면 같은 자료의 적합도가 달라진다

Industrialization1960x1 · GNP/capita1εx2 · Energy2.18εx3 · Industry1.82εDemocracy1960y1 · Press1εy2 · Opposition1.35εy3 · Elections1.04εy4 · Legislature1.3εDemocracy1965y5 · Press1εy6 · Opposition1.26εy7 · Elections1.28εy8 · Legislature1.31ε1.474[0.71, 2.24]0.864[0.64, 1.09]Direct 0.453 [0.02, 0.88]χ²(41) 72.46 · p = 0.002CFI 0.953 · RMSEA 0.102B unstandardized · first loadings fixed at 1Dashed: residual covariance · ε: indicator error

잔차 공분산 생략 · 75 countries, 11 indicators, 3 latent variables. Structural path CIs are normal expected-Fisher intervals. Good fit is not causal identification.

Political Democracy ·75countries · semopy2.3.11 / lavaan official example · exploratory model

18 / 18SEM · 2/2

18 / 18SEM · 2/2